iranian journal of economic studies vol. 4, no. 1, spring 2015,...
TRANSCRIPT
Iranian Journal of Economic Studies
Vol. 4, No. 1, Spring 2015, 101-126
Non-Linear Relationships Among Oil Price, Gold Price and Stock Market
Returns in Iran: A Multivariate Regime-Switching Approach
Siab Mamipour
Department of Economic, Kharazmi
University, Tehran,
Fereshteh Vaezi Jezeie
Department of Economic, Kharazmi
University, Tehran,
Abstract In this paper, the effects of oil and gold prices on stock market index are
investigated. We use a cointegrated vector autoregressive Markov-switching
model to examine the nonlinear properties of these three variables during the
period of January 2003 - December 2014. The Markov-switching vector-
equilibrium-correction model with three regimes representing "deep recession",
"mild recession" and "expansion" provides a good characterization of the sample
data. The results of the model show that the impact of oil price on stock returns
is positive and significant in the short run. However, it has negative effects on
stock market in the long run. Moreover, we find out that the relationship
between gold price and stock market returns varies during the period under
investigation depending on the market conditions. More specifically, the
positive gold price shock decreases the stock market returns in the short run (10
months), while it increases the stock market returns in the medium and long run.
Keywords: Stock Market Price, Oil Price, Gold Price, Markov Switching-
Vector Error Correction Model (MS-VECM), Iran.
JEL Classification: E32, E37, C32, G17.
Received: 14/10/2015 Accepted: 13/4/2016
The Corresponding Author
Iranian Journal of Economic Studies, 4(1), Spring 2015 102
1. Introduction
The stock market in economy is known as an organized and official
market of capital and its most important task is mobilization and
allocation of financial resources and turning them into capital. Achieving
the desired economic growth and development without long-run financial
resource mobilization is impossible. Yet, one of the main characteristics
of developing countries is that they suffer from fragmentation and
scarcity of savings and capital and the capital is not being conducted to
optimum routs of production. Thus, it is necessary to form a strong and
efficient capital markets in order to optimize the circulation of financial
resources and conduct them to optimum routs of economy (Dimson,
1988).
One of the important components of capital market is the stock
exchange. The existence philosophy of capital market and also stock
market is that they transfer funds from sector with surplus to funds from
sectors in need of cash. If the sectors invest the funds in producing the
capital, it is expected to be added to the wealth of society. Accordingly,
increase the returns of capital markets is the primary objective of market
policy makers and legislators (Barvch, 1988). It has been argued that the
most common method for making investment decisions in the stock
market is the study and the awareness of the stock price trend. Stock
price index shows the general trend of the stock market moves. In fact,
the degree of capital market success or failure is detected according to the
trend. Thus, portfolio managers and other natural and legal people who
deal with stock trading and other financial assets in the market need
various factors affecting the stock price under different economic
conditions, in order to maintain and increase the value of their portfolio.
Several factors affect the returns of the stock exchange. One of the
affecting factors is the oil price. Oil and its products are used as the main
sources of energy in manufacturing processes in the world; thus,
volatilities in the oil price can affect the cost of production and
profitability of production companies. Oil is the most important source of
income for some exporting countries and through this, its price and
volatilities can affect natural sector and also the capital market, so that in
many countries with poor management of oil revenues, oil price is
coupled with an increase in government revenue and monetary base and
has the inflation effects. Rising inflation will have a positive effect on the
stock price.
Non-Linear Relationships Among the Oil Price, the Gold Price … 103
Iran has 11 percent of world oil reserves and is the second largest
producer in the Organization of Petroleum Exporting Countries (OPEC).
Based on the time series data of the central bank of the Islamic republic
of Iran, 80 to 90 percent of total export revenue and 40 to 50 percent of
the annual government budget come from oil exports. In addition, the
sale of oil is over 20 percent of GDP of Iran; so Iran's economy is largely
dependent on exports of crude oil. Shocks in world oil markets can have
a great impact on the economic structure of Iran. Another factor affecting
the capital market is the price of gold. Gold is considered as the most
important world monetary standard which is mostly used in coins and
gold bullion as an international monetary reserve. Gold is also part of the
central bank's intemational reserves. From the economic point of view,
gold can be considered an strategic commodity.
Depression and expansion of stock exchange affects the national
economy. It is clear that depression and expansion of stock exchange
may be caused by many factors in the economy. As a powerful
exogenous variable, the world oil price would affect many
macroeconomic variables such as stock price index. On the other hand
the changes in the world gold price captures many international monetary
and financial development. Therefore, the main objective of this study is
to investigate the effect of the oil and the gold prices on the stock market
in Iran.
2. Literature Review
Several studies have been done on the relationships between different
markets (gold, oil and capital) which can be presented in different
categories. It has been tried to classify and present the studies based on
result analysis method, namely linear and nonlinear time-series methods.
All the studies done by Papapetrou (2001), Maghyereh (2004), Park
and Ratti (2008), Fayyad and Daly (2011) and Mohd Hussin and et al.
(2012) have used vector auto regression method (VAR) on the markets of
stock, oil and in some cases gold in different countries. The results have
been interpreted in the following.
Papapetrou (2001) showed that oil price affect real activities of
Greece’s economy and employment, while no evidence of relationship
between oil prices and stock returns can be found. Maghyereh (2004)
tested the dynamic relationship between crude oil price and stock returns
Iranian Journal of Economic Studies, 4(1), Spring 2015 104
in twenty-two emerging market economies. She showed that oil price
shocks do not have significant impact on the stock index returns.
Park and Ratti (2008) studied the oil market price volatilities on key
financial variables such as stock prices in the United States and thirteen
developed countries. The results of their studies indicate that there is
statistically a significant relationship between oil price volatility and
stock index in the mentioned markets. However, type of the relationship
depends on countries whether they are exporter or importer and this is
evident with regard to the short time period of one month. Fayyad and
Daly (2011) compared the effect of oil price shocks on the stock returns
in GCC countries, America and the UK. They have concluded that the
predictive power of oil price in the stock market has been increased after
a rise in oil price during the global crisis. Qatar and the UAE among
GCC countries and the UK compared to other countries have been more
responsive to shocks.
Mohd Hussin and et al. (2012) studied the relationship between of
strategic goods (i.e oil and gold prices) and Malaysia's stock market. The
aim of their study was to investigate the dynamic effects of oil prices and
changes in gold prices on the stock market of Malaysia by using vector
auto regression (VAR). They, applying correlation analysis, Granger
causality test, Impulse Response Function (IRF) and Variance
decomposition (VDC), showed that the stock returns do not have
integrated correlation with strategic goods in the long run. According to
Granger causality, it was observed that the two-way causality relationship
is only between stock returns and oil price. So only oil price variables
among the strategic goods will affect stock returns in the short-run in
Malaysia. This implied that the gold price is not a valid variable to
predict the stock price changes.
The non-linear methods have been used in all the studies done by
Maghyereh and Al-Kandari (2007), Arouri and Fouquau (2009), Arouri
and Nguyen (2010), Nguyen and Bhatti (2012) and Mansoor Baig and et
al. (2013), on the markets of stock, oil and in some cases gold in different
countries. The results have been interpreted in the following.
Maghyereh and Al-Kandari (2007) studied the relationship between
oil price and stock markets in Arabic countries located around the Persian
Gulf and used the non-linear correlation analysis and error correction
mechanism in their studies. They showed that in the mentioned countries,
oil price would affect the index of stock prices in the long-run but the
Non-Linear Relationships Among the Oil Price, the Gold Price … 105
relationship is non-linear. Arouri and Fouquau (2009) examined short-
run relationships between oil price and stock markets of Arabic countries
located around the Persian Gulf, by using nonparametric Kernel
regression method with the Gaussian Kernel. The result was that the
relationship between stock returns and oil price in Qatar, Oman and
United Arabic Emirates was nonlinear and there is asymmetry between
changes of oil price and stock returns in these countries. They did not
find a significant relationship in other Arabic countries in the Persian
Gulf region. Afterward in 2010, Arouri and et al. studied the stock
market responses of Arabic countries in the Persian Gulf region to
volatilities of the oil price. They showed that the stock market returns
significantly respond to oil price volatilities in Qatar, Oman, Saudi
Arabia and the United Arab Emirates. Moreover, the relationship
between the oil price and stock market in the countries is non-linear and
depend on the oil price. However, the significant relationships were not
seen between mentioned variables in Bahrain and Kuwait. Also Nguyena
and Bhatti (2012) studied the relationship between the oil price and stock
market in China and Vietnam by using copula functions and parametric
and nonparametric methods. They found that there was tail dependence
between oil world price and stock market index of Vietnam; however this
relationship had the opposite result for stock market of China. Finally,
Mansoor Baig and et al. (2013) studied the relationship between the oil
price, gold price and stock market returns of Karachi in Pakistan and
concluded that there was not a significant relationship between gold price
growth, oil price growth and stock market returns of Karachi in the long-
run.
3. Data
The monthly data index of this study is taken from Iran's stock market
and then transformed into the natural logarithms to examine the
economic relationship between the oil price (LnOil) and the gold price
(LnGold) effect on Stock market returns (LnStock). The sample size of
the study was 144 observations that were started from January 2003 until
December 2014. The study structure was a quantitative approach and
several steps were undergone before applied MS-VECM on examine the
economic relationship model.
Figure (1) shows the diagrams of logarithm values of the oil price
(OPEC basket oil (dollars)), gold price (World Gold ounces (dollars))
Iranian Journal of Economic Studies, 4(1), Spring 2015 106
and the stock price (total index of stock exchange) in Iran. As it can be
seen, the diagram related to the total index of stock has volatility and
reached its peak in December 2013 and its minimum in March 2003.The
diagram related to the oil price has a lot of volatilities and these
volatilities are more tangible during July 2008 (peak) to December 2008.
The diagram related to the gold price approximately kept its upward trend
until September 2011 and has started to decline afterwards.
Figure 1: The Variables in Level
Figure (2) shows the diagrams of logarithm values of the stock price,
oil price and gold price after taking the first differencing process. The
diagram of the stock market returns shows a lot of volatilities, as the
maximum returns were in July 2003 and the minimum was in December
2008. Also the diagram related to the oil price shows the maximum
amount of changing in June 2009 and minimum in October 2008 and
finally the diagram related to the gold price shows the maximum positive
change in August 2011 and the minimum negative change in June 2006.
Non-Linear Relationships Among the Oil Price, the Gold Price … 107
Figure 2: The Variables in First Differencing
4. Econometrics Method: MS-VECM
Various time-series models have been used to analyze the behavior of
economic and financial variables. Linear models such as Auto Regression
model (AR), Moving Averages (MA) and their combination (ARMA)
model as the common models were used. Although these linear models in
many cases are good fitting, they are not able to illustrate non-linear
dynamic patterns of variables due to their ability to detect asymmetry
(including structural break in time-series). Thus, non-linear models are
presented.
Recently, Markov Switching model (MS) has increasingly been used
in international studies and is one of the popular models of nonlinear
time-series. This model consists of multiple structures which can check
the behavior of the time-series of different regimes. A new form of
Markov Switching model is the conversion and transmission mechanisms
between different structures and regimes which are controlled by
invisible status variable which is following the Markov first order chain.
The main form of regime shift model is changing all or some of the
parameters based on Markov processes in different conditions or regimes.
Different situations are shown by invisible variable. The logic of this
kind of modeling is the combination of different distribution with
different characteristics. The current values of variables are extracted
Iranian Journal of Economic Studies, 4(1), Spring 2015 108
from this model, according to the more probable state (invisible) which is
determined by observation.
The Markov switching model was introduced by Quandt (1972),
Goldfeld and Quandt (1973) for the first time and then, was developed
for the extraction of business cycles by Hamilton (1989). Unlike other
nonlinear methods such as Artificial Neural Network (ANN) and Smooth
Transition Autoregressive (STAR) in which the transition from one
regime to another is a gradual switching, sudden switching is done in the
Markov model.
Also vector auto regression method is one of the methods used in
studying the relationships among economic variables. One of the
disadvantages of the method is that it assumes all the variables
considered in the model are stationary, but in fact they are not. When all
variables or one of them are not stationary, vector error correction model
could be a good alternative for vector auto regression model. Vector error
correction model (VECM) relates the short-run volatilities of variables to
the long-run equilibrium values and considers short-run dynamic
response of variables. Vector error correction model is a model for
linking the short-run to long-run relationships based on VAR model with
the convergence characteristics.
When the series are non-stationary in levels and cointegrated, a
bivariate vector error correction model (VECM) of order p can be used to
jointly modeling rates behavior. Formally: 1
1
1
p
t i t i t t
i
y y y
(1)
Where ty =[LnStock LnOil LnGold]´ is a three-dimensional vector of
Stock, Oil and Gold Market; LnStock is the natural logarithm of total
price index differential as stock market return; LnOil represents the oil
price differential and LnGold is the gold price differential. i is 3×3
autoregressive short-run parameters matrices and is the long-run
impact matrix, whose rank r determines the number of cointegrating
vectors. In the bivariate case, can be partitioned into a 3×1 vector
of long-coefficients, representing the long-run relationships among the
variables, and a 3×1 vector α containing the equilibrium correction
(speed of adjustment) coefficients ( ).
Non-Linear Relationships Among the Oil Price, the Gold Price … 109
Model (1) assumes that the relationships among stock market, gold
and oil price rates are symmetric and linear. Stock market return series
are characterized by occasional jumps or structural changes in their levels
or volatility. The presence of important discrete economic events induces
substantial non-linearity in the stochastic process and distorts inference if
it is not appropriately modeled. However, most of the studies adopt a
deterministic approach, which consists identifying structural breaks in the
series and modeling these shifts by augmenting the empirical
specification with an appropriate set of dummy variables or by
conducting split sample analyses (Blot and Labondance, 2011;
Panagopoulos and Spiliotis, 2012). When regime shifts are stochastic
rather than deterministic, these approaches may lead to biased or at least
inefficient results (Dahlquist and Gray, 2000; Clarida et al., 2006). In
such cases, a multivariate generalization of the univariate Markov-
switching (MS) model proposed by Hamilton(1989) represents a viable
alternative to allow for stochastic behavioural changes.
Multivariate generalized Markov switching model is introduced by
Krolzig (1996) as Markov Switching-Vector Error Correction Model
(MS-VECM) and is related to the concept of multiple equilibriums in
dynamic economic theory.
According to Krolzig study (1996), a Markov Switching-Vector Error
Correction Model (MS-VECM) in the form of equation (2) is a model
with regime dependent parameters:
1
1
1
( ) ( ) ( )p
t t t t i t t k t
k
y v s s x t s y u
(2)
The model is related to the concept of multiple equilibriums in
dynamic economic theory. From now on, every regime which is defined
by a system with the phrase ( )ts and long-run equilibrium ( )ts , would
be determined in the form of equation (3):
1
1
1
( ) ( ) ( )p
t t t t i t k t t
k
y s y s t y s u
(3)
In the model, it is assumed that a regime which is occurs in t time, is
invisible and depends on ( )ts an invisible process.
To complete the model, characteristics of ts process should be
identified. ts in the Markov switching model is considered as a first
Iranian Journal of Economic Studies, 4(1), Spring 2015 110
degree Markov process. This assumption reflects the fact that ts only
depends on last period regime, means 1ts . In the following, we would
complete the model by introducing the equation (4) the transition
probabilities from one state to another one:
1 2 1 2 1Pr , ,...; , ,... Prt t t t t t t ijs j s i s k y y s j s i p (4)
Transition between regimes can be shown by transition probability
matrix. For example, the matrix is in the form of equation (5) for a two-
regime model:
1 1 11 12
1 1 21 22
Pr( 1 1) Pr( 2 1)
Pr( 1 2) Pr( 2 2)
t t t t
t t t t
s s s s p pP
s s s s p p
(5)
In equation (5), ( , 1,2)ijp i j shows the probability of transition from
ts j , assuming 1ts i and 1 2 1i ip p .
The MS-VECM model allows all the variables in it to be shocked, in
the context of the model. When the system is affected by the regime
transition, impulse response is checked by considering the transition of
variables of dependent on VAR feature regime and hidden Markov chain.
Impulse responses are often used in examining the relationship between
the dynamic model variables. So that the ultimate impact of impulse on
the system is being followed over time.
In Krolzig and Toro (1999) study, at first MS (M) -VECM(P-1) model
is considered in the form of equation (6) to analysis the impulse response
in Markov switching model:
1 1 1 1 1...t t t t p t p ty M y y y u (6)
In equation (6): 1:... :M M . MS (M)-VAR(P) model corresponded
to equation (6) is in the form of equation (7).
1 1t t t p t p ty M A y A y u (7)
In equation (7), 1j j jA and 1 1kA I for
1j j jA and 1< j < p are equal to 0p k . A show of MS(M)-
VAR(1) accumulated with a process of MS(M)-VAR(P) is used to
calculate the impulse response function. By considering
1( ,..., )t t t py y y equation (7) is rewritten as equation (8):
Non-Linear Relationships Among the Oil Price, the Gold Price … 111
1t t t ty H JAy u (8)
The state-space representation is completed in the form of equation (9)
by the VAR(1) representation of the Markov chain (Hamilton, 1994):
1t t tF v (9)
In equation (9), t is the unobservable ( 1)M state vector consisting
of the indicator variables ( )tI S m for 1,...,m M ; Hence the
expectation of t hy conditional upon 1{ , , }t t tu y is given by equation
(10):
1t h t t h t t h ty H JAy
)10)
In equation (10), conditional expectation of t h is as equation (11):
(11)
According to equation (11), impulse response analyses (response of
system to normal shifts in variables) of linear VAR models are as
equation (12):
ht hj
jt
yJA
u
(12)
In equation (12), j is the jth column of the identity matrix. If
variance-covariance matrix ∑u is regime-dependent the standardized and
orthogonalized impulse-responses also become regime-dependent:
( )ht ht j
jt
yJA D
u
(13)
In equation (13), ( )t t tu D
and ( )tD
is a lower triangular matrix
resulting from Choleski decomposition of ( ) ( ) ( )t t tu D D
5. Empirical Results and Discussion
5.1. Unit Root Test At first, stationary of variables should be ensured to avoid spurious
regression. In this section, stationary of variables are checked by the
Augmented Dickey Fuller (ADF), Dickey Fuller GLS (ERS) and
Kwiatkowski-Philips-Schmidt-Shin (KPSS) tests. ADF and ERS tests
showed that under the null hypothesis of a unit root; their outputs
reported MacKinnon lower-tail critical and p-values for these tests, but
Iranian Journal of Economic Studies, 4(1), Spring 2015 112
the KPSS test differed from the other unit root tests described here in that
the series are assumed to be stationary under the null hypothesis and the
KPSS output only provided the asymptotic critical values tabulated by
KPSS.
Test results, presented in Table 1, clearly indicate the presence of a
unit root for all the series in levels and a rejection for the series in first-
differences, providing evidence of an I(1) behavior.
Table 1: Unit Root Test
Ln(gold) Ln(oil) Ln(stock)
First diff. Level First diff. Level First diff. Level
a)Unit root test -10.6824
(0.000)
-1.6817
(0.4384) -7.5060
(0.000)
-2.0693
(0.2574)
-6.2123
(0.000)
-0.6374
(0.8574)* ADF
-10.7053
(0.000)
0.7368
(0.4624) -7.2494
(0.000)
-1.0174
(0.3107)
-6.0998
(0.000)
0.8738
(0.3837) DF-GLS
b) Stationary test
0.380 1.3077 0.1684 1.2106 0.1674 1.1160 KPSS** *The value in parentheses is represented to the p-value.
** Asymptotic critical values for the KPSS test are 0.739, 0.463 and 0.347 at the 1,
5 and 10% levels, respectively.
5.2. Cointegration Test
To estimate the cointegration model of Johansen, The optimal interval of
variables must be provided at first. Thus LR test and information
criterion are used to identify the optimal interval and also the optimal
number of regime and optimal model in Markov switching model. The
model is optimal when the amount of its information criterion is
minimum. Three of the most common information criterion includes:
The Akaike Information Criterion (AIC) Test:
2 2log( )AIC k L (14)
The Schwarz Information Criterion (SC) Test:
ln( ) 2 log( )SC k n L (15)
The Hannan-Quinn Information Criterion (HQ) Test:
2 ln(ln( )) 2 ln( )HQ k n l (16)
Where k represents the number of parameters, n is the number of
observations and L is the maximized likelihood value.
Non-Linear Relationships Among the Oil Price, the Gold Price … 113
To indicate the optimal lag of VAR model and according to the table
(2), the number 8 is considered as the maximum of optimal interval in the
model. The optimal interval is 2, according to the existed criteria.
Table 2: VAR Lag Order Selection Criteria
Lag LogL LR FPE AIC SC HQ
0 -181.3006 NA 0.003017 2.710303 2.774552 2.736412
1 616.0590 1547.816 2.78e-08 -8.883220 -8.626221 -8.778782
2 650.5638 65.45768* 1.91e-08* -9.258291* -8.808543* -9.075525*
3 659.4575 16.47950 1.92e-08 -9.256728 -8.614231 -8.995633
4 667.0748 13.77844 1.96e-08 -9.236395 -8.401148 -8.896972
5 669.6355 4.518813 2.16e-08 -9.141699 -8.113703 -8.723947
6 678.255 14.83151 2.17e-08 -9.136111 -7.915366 -8.640031
7 685.1897 11.62494 2.25e-08 -9.105731 -7.692237 -8.531323
8 692.8311 12.47349 2.30e-08 -9.085752 -7.479508 -8.433015
* indicates lag order selected by the criterion, LR: sequential modified LR test
statistic (each test at 5% level), FPE: Final prediction error
At this point, we determine the number of cointegration vectors. This
is a means of testing for cointegration in a multivariate context, where
there is the possibility of more than one cointegrating vector being
present. It follows the same principles as the Engle-Granger approach to
cointegration, in so far as the order of integration of the variables are first
assessed, if the variables are I(1) the Johansen Maximum Likelihood
(ML) procedure can then be used to determine whether a stable long-run
relationship exists between the variables.
Johansen cointegration method based on ˆtrace , trace test and max̂ ,
maximum eigenvalue test is used to determine the number of
cointegration vectors. Summary of the results of maximum eigenvalue
test and trace test is in table (3). According to results of Table (3), all two
statistics of ˆtrace and max̂ confirm the existence of a cointegration
vector; because null hypothesis of no convergence vector is ruled out by
both statistics but the hypothesis of presence of more than one integration
vector is not ruled out.
Iranian Journal of Economic Studies, 4(1), Spring 2015 114
Table 3: Johansen Test Outputs
Hypothesized
No. of CE(s) Trace statistics
Maximum
Eigenvalue statistics
ˆtrace
5% critical
value max̂ 5% critical
value
None* 41.2697*
(0.0098) 35.1927
25.6711*
(0.0163) 22.2996
At most 1 15.5985
(0.1940) 20.2618
12.1865
(0.1754) 15.8921
At most 1 3.4119
(0.5067) 9.1645
3.4119
(0.5067) 9.1645
The value in parentheses is represent the p-value
* denotes rejection of the hypothesis at the 0.05 level
Table 4 shows the convergent and normalized vector of the model. As
it is clear, the cointegration vector under consideration is normalized with
regard to variable of Lnstock.
Based on the results of linear error correction, stock price index has a
reverse relationship with the oil price and a direct relationship with the
gold price in long-run. Since, the variables used in this model are
logarithmic; we can interpret coefficients as elasticity. The long-run
elasticity of stock price index with regard to the oil price is -5.2399, this
means that one percent increase in the oil price decreases the stock price
index up to 5.2399 percent. Also, the long-run elasticity of the stock price
index with regard to the gold price is 4.5395 which means that one
percent increase in the gold price increases the stock price index up to
4.5395 percent in the long-run.
On the one hand, the oil price increases along with the increase in the
oil incomes which disrupts the allocation of financial sources in
government sector and will develop marginal investments and unfinished
projects; on the other hand, increase in importing consumer would
decrease the competition power of internal products and output of private
sector investments. As a result, those who work in private sectors are not
motivated to invest in exchangeable products sector and this issue affects
the capital market. The direct relation between the stock price and the
gold price has somewhat encountered difficulty in the long-run.
Considering such results, we should take necessary caution. Such
unexpected result may be due to stock market inefficiency and lack of
Non-Linear Relationships Among the Oil Price, the Gold Price … 115
currency basket among people. Other reason of such results may be the
fact that the gold price in Iran is affected by two sides. On one hand, it is
affected by world gold price and on the other hand, it is affected by the
assessed economic policies.
Table 4: Cointegration Vector of Linear
CoinEq1 Cointegrating Eq:
1.0000 Lnstock(-1)
5.2399
(1.4500)
[3.6135] Lnoil(-1)
-4.5395
(1.2388)
[-3.6643] Lngold(-1)
-2.2744
(3.8802)
[-0.5861] C
The value in () is represent to the standard errors
The value in [ ] is represent to the t-statistics
5.3. MS-VECM Result
5.3.1. Testing for Non-Linear Relations and Regime Characteristics
Markov Switching models are created through change in mean, intercept,
heteroskedasticity and autoregression coefficients. Two conditions are
required for selecting an optimum model. Firstly, the null hypothesis
which is based on stable regime in the model must be rejected. Secondly,
the mentioned model must be more appropriate in terms of information
criteria than the probable models in which the first condition is proved.
Through investigating various techniques, taking into consideration
the nature of data and the optimum interval, we could identify three
regimes. Then, the models were compared based on AIC, SC, and HQ
criteria and LR test. At the end, MSIH(3)-VECM(1) model was selected
as the superior one. In this model, intercept and heteroskedasticity
variance depend on the regime.
Table 5 shows the results of estimating parameters of the above model
by using maximum likelihood method. The statistical amount of LR test
based on the linearity behavior of variables is 78.7059 and based on the
Iranian Journal of Economic Studies, 4(1), Spring 2015 116
p-value number related to Davises Statistics, this hypothesis is rejected
and the non-linearity relation between these variables is confirmed.
Table 5: Criterion Test Results on the Markov Switching Models
MSIH(3) – VECM(1) Linear VECM
Log-likelihood 706.626 667.2740
AIC criterion -9.3187 -9.1025
HQ criterion -8.9380 -8.9248
SC criterion -8.3820 -8.6653
LR linearity test: 78.7059 DAVIES=[0.0000]
Chi(18) =[0.0000] Chi(24)=[0.0000]
The results of estimating MSIH(3)-VECM(1) model for investigating
the impact of the stock market return on the oil and the gold prices are
presented in Table 6. As it shows, during this study the impact of these
parameters can be divided into three regimes and also the variable
coefficients are statistically significant. Since, the intercept of regime 1 is
less than regimes 2 and 3, we can say that stock output in regime 1 is less
than the two other regimes. So, regime 1 is considered as deep recession,
regime 2 as less recession and regime 3 as expansion situation.
Coefficients of variables are significant. As it can be seen in Table 6,
the sign related to independent variable, oil price, is positive which
shows that there was a direct relation between the oil price and the stock
output in the short run. Moreover, gold price has negative effect on the
stock market returns.
CYN coefficient which shows the adjustment rate demonstrates that
for the total index of stock exchange in each period 0.6 percent is
adjusted which indicates the low rate of adjustment in the model.
The results in Figure (3) show that the three-regime MS-VECM model
is suitable for estimating the variables in the financial relationship model
because the fitted and 1-step predicted probabilities for all variables in
the system fit to the mean.
Non-Linear Relationships Among the Oil Price, the Gold Price … 117
Table 6: MSIH(3)-VECM(1) Outputs
DLN(stock) t DLN(oil) t DLN(gold) t
Intercepts
Const (Regime 1)
Const (Regime 2)
Const (Regime 3)
-0.0523
(-6.4493)
-0.0077
(-1.5396)
0.0546
(5.9902)
-0.1248
(-4.0419)
0.0015
(0.1549)
0.0021
(0.0902)
-0.0179
(-1.0383)
0.0042
(0.7935)
0.0193
(1.5447)
Autoregressive coefficients
DLN(stock) t-1
DLN(oil) t-1
DLN(gold) t-1
0.3017
(4.5659)
0.0513
(1.2705)
-0.2335
(-3.5185)
0.0807
(0.5100)
0.1970
(2.0761)
-0.0004
(-0.0035)
-0.0557
(-0.6108)
0.0263
(0.5794)
0.1265
(1.4918)
Adjustment coefficient
CYN -0.006
(-2.7001)
-0.0124
(-2.6621)
-0.0013
(-0.5220)
Standard Error
SE (Regime 1)
SE (Regime 2)
SE (Regime 3)
0.0266
0.0241
0.0389
0.0922
0.0731
0.0400
0.0603
0.0336
0.0406 The value in () is represent to the t-statistics
Iranian Journal of Economic Studies, 4(1), Spring 2015 118
Figure 3: MSIH(3)-VECM(1) Fit
Based on the results in Table 7, the dominant economical regime is
the second regime so that markets under investigation are in normal
conditions or recession (regime 2) with the probability of 0.5756 and the
average survival duration of 4.04, in prosperity duration (regime 3) with
the probability of 0.326 and the average survival duration of 2.38, and in
slump situation (regime 1) with the probability of 0.098 and the average
survival duration of 1.63.
Table 7: Regime Properties
No. of observations Probability Duration
Regime 1 14.3 0.0984 1.63
Regime 2 81.4 0.5756 4.04
Regime 3 46.4 0.3260 2.38 The situations of the three regimes are presented in Figure 4.
Non-Linear Relationships Among the Oil Price, the Gold Price … 119
Figure 4: MSIH(3)-VECM(1) Probabilities Sketched
In order to investigate the degree of regimes instability and
probabilities of transfer of each regime to another, we have extracted
Matrix of transition probability. As Table 8 indicates, the probability of
transfer from regime one (deep recession) to regime 2 (less recession) is
0.61 and from regime 2 to regime 1 is 0.009. Probability values show that
regime 2 is more stable than regime 1. Also, the probability of transfer
from regime 1 to regime 3 (expansion) is very low and from regime 3 to
regime 1 is 0.168. As a result, regime 1 is more stable than regime 3. At
most, the probability of move from regime 2 to regime 3 is 0.237 and
from regime 3 to regime 2 is 0.251. So, the stability of regimes 2 and 3 is
almost the same.
In addition, the probability of transfer from regime 2 to regime 3 is
0.75 which is more stable than the other ones. The results obtained from
regimes transfer is in line with economic cycle’s theories and are
expected so that change of stagnation condition into prosperity condition
happens with less probability and in the long run, while change of
prosperity condition into stagnation condition happens with more
probability and in a short run.
Iranian Journal of Economic Studies, 4(1), Spring 2015 120
Table 8: Matrix of Transition Probabilities
Regime 1 Regime 2 Regime 3
Regime 1 0.3860 0.6140 1.846e-006
Regime 2 0.0096 0.7526 0.2377
Regime 3 0.1683 0.2515 0.5803
5.3.2. Impulse Response Analysis
In next stage, the impact of shock on the particular variable on other
variables is investigated through Impulse Response Function (IR). The
analysis of active reciprocal impacts of created impulses in the system is
done through variance decomposition and Impulse Response Functions.
Impulse Response Functions (IRF) shows the active behavior of
pattern variables at the time of creating impulses on each of the pattern
variables in the duration. These impulses are usually selected as one
standard deviation. The origin or the beginning point of response variable
movement is the values related to the stable situation of system (without
impulse).
Figure 5 shows the response of the stock price index, the oil price and the
gold price toward one shock standard deviation to each variable in
different three regimes.
The response of stock market to the shock on the gold price in each of
the three regimes is similar and it is negative in short time, also it
decreases the stock price. But, gradually the effects disappear and after
about 10 months, the gold price shock would have a positive effect on the
stock market and it would have a positive and stable long term effect.
Moreover, the results show that the response of stock market to the
positive shock on the oil price in each of the situations decreases the
stock price. This effect is stable in all the short term and long term
durations.
Non-Linear Relationships Among the Oil Price, the Gold Price … 121
Figure 5 : Impulse Response Analysis
6. Conclusion In this study, the impact of oil and gold prices on Iran's stock market
were investigated through monthly data covering the period January 2003
to December 2014. A cointegrated vector autoregressive Markov
switching model was used to examine the nonlinear properties of the
variables. Impact of shocks on each variable was investigated and was
anticipated.
Three different regimes including “deep recession”, “mild recession”
and “expansion” are used to characterize the sample data respectively.
The results obtained from the model show that the impact of the oil price
on stock return in each regime is statistically significant and positive in
the short run. However, it has negative effect on stock market in the long
run.
The relation between the gold price and the stock market return varies
during the period under investigation depending on the market
conditions. More specifically, the positive gold price shock decreases the
stock market returns in the short time (10 month), while it increases the
stock market return in the medium and long run. Our finding might have
implications for policymakers and investors in the stock market of Iran.
Iranian Journal of Economic Studies, 4(1), Spring 2015 122
References Arouri, M. E. H., & Fouquau, J. (2009). On the short-term influence of
oil price changes on stock markets in GCC countries: linear and
nonlinear analyses, Cornell University Library.
Arouri, M. E. H., & Nguyen, D. K. (2010). Oil prices, stock markets and
portfolio investment: Evidence from sector analysis in Europe over
the last decade. Energy Policy, 38(8), 4528–4539.
Barvch, L. (1988). Toward a Theory of Equitable and Efficient
Accounting policy. The Accounting Review, 63, 15-37.
Blot, C., & Labondance, F. (2011). Bank interest rate pass-through in the
euro zone: monetary policy transmission during the boom and since
the financial crash. Paper presented at the 15th Annual Conference on
Macroeconomic Analysis and International Finance.
Clarida, R. H., Sarno, L., Taylor, M. P., & Valente, G. (2006). The role
of asymmetries and regime shifts in the term structure of interest rates.
Journal of Business, 79(3), 1193-1224.
Dahlquist, M., & Gray, S. F. (2000). Regimes switching and interest rates
in the European monetary system. Journal of International
Economics, 50, 399-419.
Dimson, E., & Mussavian, M. (1998). A brief history of market
efficiency. European Financial Management, 4, 91-193.
Engle, R F., & Granger, C.W. J. (1987). Cointegration and error
correction: representation, estimation and testing. Econometrica, 55,
251-276.
Fayyad, A., & Daly, K. (2011). The impact of oil price shocks on stock
market returns: comparing GCC countries with the UK and USA.
Emerging Markets Review, 12(1), 61-78.
Goldfeld, S. M., & Quandt, R. E. (1973). A Markov model for switching
regressions. Journal of Econometrics, 1(1), 3-15.
Hamilton J. D. (1989). A New Approach to the Economic Analysis of
Nonstationary Time Series and the Business Cycle. Econometrica, 57,
357-384.
Hamilton, J. D. (1994). Time series analysis (Vol. 2). Princeton:
Princeton university press.
Krolzig, H. M. (1996). Statistical analysis of cointegrated VAR processes
with Markovian regime shifts. Berlin: Humboldt University Press.
Non-Linear Relationships Among the Oil Price, the Gold Price … 123
Krolzig, H. M., & Toro, J. (1999). A new approach to the analysis of
shocks and the cycle in a model of output and employment. European
University Institute. Working paper ECO, no. 99-30.
Maghyereh, A. (2004). Oil price shocks and emerging stock market: a
generalized VAR approach. International Journal of Applied
Econometrics and Quantitative Studies, 1(2), 27-40.
Maghyereh, A., & Al-Kandari, A. (2007). Oil prices and stock markets in
GCC countries: new evidence from nonlinear cointegration analysis.
Managerial Finance, 33(7), 449-460.
Mansoor Baig, M., Shahbaz, M., Imran, M., Jabbar, M., & Ain, Q.U.
(2013). Long run relationship between gold prices, oil prices and
Karachi stock market. Acta Universitatis Danubius, 9(5), 28-39.
Mohd Hussin, M. Y., Muhammad, F., Abu, M. F., & Awang, S. A.
(2012). Macroeconomic Variables and Malaysia Islamic Stock
Market: A Time Series Analysis. Journal of Business Studies
Quarterly, 3(4), 1-13.
Mohd Hussin, M. Y., Muhammad, F., Abu-Hussin, M. F., & Abdul
Razak, A. (2012). The Relationship between Oil Price, Exchange Rate
and Islamic Stock Market in Malaysia. Research Journal of Finance
and Accounting, 3(5), 83-92.
Nguyen, C.C., & Bhatti, M.I. (2012). Copula model dependency between
oil prices and stock markets: Evidence from China and Vietnam.
Journal of International Financial Markets, Institutions & Money, 22,
758–773.
Panagopoulos, Y., & Spiliotis, A. (2012). Is the Eurozone homogeneous
and symmetric? An interest rate pass-through approach before and
during the recent financial crisis. Paper presented at the 15th Annual
Conference on Macroeconomic Analysis and International Finance.
Park, J., & Ratti, R. A. (2008). Oil price shocks and stock markets in the
U.S. and 13 European countries, Energy Economics, 30(5), 2587-
2608.
Papapetrou, E. (2001). Oil price shocks, stock market, economic activity
and employment in Greece. Energy Economics, 23(5), 511-532.
Quandt, R. E. (1972). A New Approach to Estimating Switching
Regressions. Journal of the American Statistical Association, 67(338),
306-310.
Iranian Journal of Economic Studies, 4(1), Spring 2015 124
Appendix:
A1: Regime Classification
Regime 1 Regime 2 Regime 3
2003:3 - 2003:3 [0.9756]
2003:9 - 2003:9 [0.9990]
2004:2 - 2004:2 [0.6931]
2006:6 - 2006:6 [0.9874]
2008:8 - 2008:12
[0.9996]
2009:12 - 2009:12
[0.9029]
2011:5 - 2011:5 [0.9693]
2011:10 - 2011:11
[0.7666]
2014:12 - 2014:12
[1.0000]
2003:4 - 2003:4 [0.9999]
2003:10 - 2003:10
[0.9979]
2004:3 - 2004:6 [0.9544]
2004:8 - 2005:11
[0.9528]
2006:1 - 2006:4 [0.9546]
2006:7 - 2007:8 [0.9118]
2007:11 - 2007:12
[0.9547]
2008:2 - 2008:5 [0.8729]
2009:1 - 2009:7 [0.9363]
2010:1 - 2010:2 [0.9769]
2010:5 - 2010:6 [0.8638]
2010:10 - 2010:12
[0.8245]
2011:6 - 2011:7 [0.8996]
2011:12 - 2012:8
[0.8976]
2013:1 - 2013:3 [0.8131]
2014:1 - 2014:11
[0.8463]
2003:5 - 2003:8
[0.9993]
2003:11 - 2004:1
[0.9321]
2004:7 - 2004:7
[0.9884]
2005:12 -
2005:12 [0.9806]
2006:5 - 2006:5
[0.9700]
2007:9 - 2007:10
[0.8875]
2008:1 - 2008:1
[0.9957]
2008:6 - 2008:7
[0.9999]
2009:8 - 2009:11
[0.8971]
2010:3 - 2010:4
[0.8850]
2010:7 - 2010:9
[0.9069]
2011:1 - 2011:4
[0.9037]
2011:8 - 2011:9
[0.9565]
2012:9 - 2012:12
[0.8980]
2013:4 - 2013:12
[0.9153]
Non-Linear Relationships Among the Oil Price, the Gold Price … 125
A2: MSIH(3)-VECM(1) Diagrams
Iranian Journal of Economic Studies, 4(1), Spring 2015 126
A3: MSIH(3)-VECM(1) Residuals
A4: MSIH(3)-VECM(1) Regime Shifts